Context-Free Grammars

Report
Context-Free Grammars
Formalism
Derivations
Backus-Naur Form
Left- and Rightmost Derivations
1


Madam I’m Adam

Able was I ere I saw Elba

madamimadam ablewasIereIsawelba

w=wR

Not a regular language.

Think about 0n10n , Pumping Lemma
Informal Comments



A context-free grammar is a notation for describing
languages.
It is more powerful than finite automata or RE’s,
but still cannot define all possible languages.
Useful for nested structures, e.g., parentheses in
programming languages.
3

Informal Comments – (2)




Basic idea is to use “variables” to stand for sets
of strings (i.e., languages).
These variables are defined recursively, in terms
of one another.
Recursive rules (“productions”) involve only
concatenation.
Alternative rules for a variable allow union.
4

Example: CFG for { 0n1n | n > 1}

Productions:
S -> 01
S -> 0S1


Basis: 01 is in the language.
Induction: if w is in the language, then so is 0w1.
5

CFG Formalism



Terminals = symbols of the alphabet of the
language being defined.
Variables = nonterminals = a finite set of other
symbols, each of which represents a language.
Start symbol = the variable whose language is the
one being defined.
6

Productions


A production has the form variable (head) -> string
of variables and terminals (body).
Convention:





A, B, C,… and also S are variables.
a, b, c,… are terminals.
…, X, Y, Z are either terminals or variables.
…, w, x, y, z are strings of terminals only.
, , ,… are strings of terminals and/or variables.
7

Example: Formal CFG





Here is a formal CFG for { 0n1n | n > 1}.
Terminals = {0, 1}.
Variables = {S}.
Start symbol = S.
Productions =
S -> 01
S -> 0S1
8

Derivations – Intuition

We derive strings in the language of a CFG by
starting with the start symbol, and repeatedly
replacing some variable A by the body of one of its
productions.

That is, the “productions for A” are those that have
head A.
9

Derivations – Formalism



We say A =>  if A ->  is a production.
Example: S -> 01; S -> 0S1.
S => 0S1 => 00S11 => 000111.
10

Iterated Derivation



=>* means “zero or more derivation steps.”
Basis:  =>*  for any string .
Induction: if  =>*  and  => , then  =>* .
11

Example: Iterated Derivation



S -> 01; S -> 0S1.
S => 0S1 => 00S11 => 000111.
Thus S =>* S; S =>* 0S1; S =>* 00S11; S =>*
000111.
12

Sentential Forms


Any string of variables and/or terminals derived
from the start symbol is called a sentential form.
Formally,  is a sentential form iff
S =>* .
13

Language of a Grammar

If G is a CFG, then L(G), the language of G, is {w |
S =>* w}.
Example: G has productions S -> ε and S -> 0S1.

L(G) = {0n1n | n > 0}.

14

Context-Free Languages




A language that is defined by some CFG is called
a context-free language.
There are CFL’s that are not regular languages,
such as the example just given.
But not all languages are CFL’s.
Intuitively: CFL’s can count two things, not
three.
15

BNF Notation



Grammars for programming languages are often
written in BNF (Backus-Naur Form ).
Variables are words in <…>; Example:
<statement>.
Terminals are often multicharacter strings
indicated by boldface or underline; Example:
while or WHILE.
16

BNF Notation – (2)


Symbol ::= is often used for ->.
Symbol | is used for “or.”


A shorthand for a list of productions with the same left
side.
Example: S -> 0S1 | 01 is shorthand for S -> 0S1
and S -> 01.
17

BNF Notation – Kleene Closure



Symbol … is used for “one or more.”
Example: <digit> ::= 0|1|2|3|4|5|6|7|8|9 <unsigned
integer> ::= <digit>…
Translation: Replace … with a new variable A and
productions A -> A | .
18

Example: Kleene Closure

Grammar for unsigned integers can be replaced by:
U -> UD | D
D -> 0|1|2|3|4|5|6|7|8|9
19

BNF Notation: Optional Elements



Surround one or more symbols by […] to make
them optional.
Example: <statement> ::= if <condition> then
<statement> [; else <statement>]
Translation: replace [] by a new variable A with
productions A ->  | ε.
20

Example: Optional Elements
Grammar for if-then-else can be replaced by:
S -> iCtSA
A -> ;eS | ε

21

BNF Notation – Grouping

Use {…} to surround a sequence of symbols that
need to be treated as a unit.


Typically, they are followed by a … for “one or
more.”
Example: <statement list> ::= <statement>
[{;<statement>}…]
22

Translation: Grouping



Create a new variable A for {}.
One production for A: A -> .
Use A in place of {}.
23

Example: Grouping

L -> S [{;S}…]
Replace by L -> S [A…]


A stands for {;S}.
Then by L -> SB B -> A… | ε A -> ;S


A -> ;S
B stands for [A…] (zero or more A’s).
Finally by L -> SB
A -> ;S

B -> C | ε
C -> AC | A
C stands for A… .
24

Leftmost and Rightmost Derivations

Derivations allow us to replace any of the
variables in a string.


Leads to many different derivations of the same string.
By forcing the leftmost variable (or alternatively,
the rightmost variable) to be replaced, we avoid
these “distinctions without a difference.”
25

Leftmost Derivations


Say wA =>lm w if w is a string of terminals only
and A ->  is a production.
Also,  =>*lm  if  becomes  by a sequence of 0
or more =>lm steps.
26

Example: Leftmost Derivations




Balanced-parentheses grammmar: S -> SS | (S) | ()
S =>lm SS =>lm (S)S =>lm (())S =>lm (())()
Thus, S =>*lm (())()
S => SS => S() => (S)() => (())() is a derivation,
but not a leftmost derivation.
27

Rightmost Derivations


Say Aw =>rm w if w is a string of terminals
only and A ->  is a production.
Also,  =>*rm  if  becomes  by a sequence of 0
or more =>rm steps.
28

Example: Rightmost Derivations




Balanced-parentheses grammmar: S -> SS | (S) | ()
S =>rm SS =>rm S() =>rm (S)() =>rm (())()
Thus, S =>*rm (())()
S => SS => SSS => S()S => ()()S => ()()() is
neither a rightmost nor a leftmost derivation.
29

Parse Trees
Definitions
Relationship to Left- and Rightmost
Derivations
Ambiguity in Grammars
30

Parse Trees

Parse trees are trees labeled by symbols of a
particular CFG.
Leaves: labeled by a terminal or ε.

Interior nodes: labeled by a variable.



Children are labeled by the body of a production for
the parent.
Root: must be labeled by the start symbol.
31

Example: Parse Tree
S -> SS | (S) | ()
S
S
(
S
S
(
)
(
)
)
32

Yield of a Parse Tree

The concatenation of the labels of the leaves in leftto-right order


That is, in the order of a preorder traversal.
is called the yield of the parse tree.
Example: yield of
is (())()
S
S
(
S
S
(
)
(
)
)
33

Generalization of Parse Trees


We sometimes talk about trees that are not exactly
parse trees, but only because the root is labeled by
some variable A that is not the start symbol.
Call these parse trees with root A.
34

Parse Trees, Leftmost and Rightmost
Derivations
Trees, leftmost, and rightmost derivations
correspond.
We’ll prove:


1.
2.
If there is a parse tree with root labeled A and yield w,
then A =>*lm w.
If A =>*lm w, then there is a parse tree with root A and
yield w.
35

Proof – Part 1




Induction on the height (length of the longest path
from the root) of the tree.
Basis: height 1. Tree looks like
A -> a1…an must be a production.
A
Thus, A =>*lm a1…an.
...
a1
an
36

Part 1 – Induction


Assume (1) for trees of height < h, and let this
tree have height h:
By IH, Xi =>*lm wi.


Note: if Xi is a terminal, then Xi = wi.
Thus, A =>lm X1…Xn =>*lm w1X2…Xn =>*lm
w1w2X3…Xn =>*lm … =>*lm w1…wn.
X1
w1
A
...
Xn
wn
Proof: Part 2


Given a leftmost derivation of a terminal string, we
need to prove the existence of a parse tree.
The proof is an induction on the length of the
derivation.
38

Part 2 – Basis

If A =>*lm a1…an by a one-step derivation, then
there must be a parse tree
A
a1
...
an
39

Part 2 – Induction



Assume (2) for derivations of fewer than k > 1
steps, and let A =>*lm w be a k-step derivation.
First step is A =>lm X1…Xn.
Key point: w can be divided so the first portion is
derived from X1, the next is derived from X2, and
so on.

If Xi is a terminal, then wi = Xi.
40

Induction – (2)

That is, Xi =>*lm wi for all i such that Xi is a
variable.



And the derivation takes fewer than k steps.
By the IH, if Xi is a variable, then there is a parse
tree with root Xi and yield wi.
Thus, there is a parse tree
A
X1 . . .
Xn
w1
wn
41

Parse Trees and Rightmost Derivations


The ideas are essentially the mirror image of the
proof for leftmost derivations.
Left to the imagination.
42

Parse Trees and Any Derivation



The proof that you can obtain a parse tree from a
leftmost derivation doesn’t really depend on
“leftmost.”
First step still has to be A => X1…Xn.
And w still can be divided so the first portion is
derived from X1, the next is derived from X2, and
so on.
43

Ambiguous Grammars



A CFG is ambiguous if there is a string in the
language that is the yield of two or more parse trees.
Example: S -> SS | (S) | ()
Two parse trees for ()()() on next slide.
44

Example – Continued
S
S
S
S
S
(
S
)
(
(
)
S
)
(
S
S
)
(
S
)
(
)
45

Ambiguity, Left- and Rightmost Derivations



If there are two different parse trees, they must
produce two different leftmost derivations by the
construction given in the proof.
Conversely, two different leftmost derivations
produce different parse trees by the other part of
the proof.
Likewise for rightmost derivations.
46

Ambiguity, etc. – (2)
Thus, equivalent definitions of “ambiguous
grammar” are:

1.
2.
There is a string in the language that has two different
leftmost derivations.
There is a string in the language that has two different
rightmost derivations.
47

Ambiguity is a Property of
Grammars, not Languages

For the balanced-parentheses language, here is
another CFG, which is unambiguous.
B -> (RB | ε
B, the start symbol,
derives balanced strings.
R -> ) | (RR
R generates certain strings
that have one more right
paren than left.
48

Example: Unambiguous Grammar
B -> (RB | ε

R -> ) | (RR
Construct a unique leftmost derivation for a given
balanced string of parentheses by scanning the string
from left to right.

If we need to expand B, then use B -> (RB if the next
symbol is “(”; use ε if at the end.

If we need to expand R, use R -> ) if the next symbol is
“)” and (RR if it is “(”.
49

The Parsing Process
Remaining Input:
(())()
Steps of leftmost
derivation:
B
Next
symbol
B -> (RB | ε
R -> ) | (RR
50

The Parsing Process
Remaining Input:
())()
Steps of leftmost
derivation:
B
(RB
Next
symbol
B -> (RB | ε
R -> ) | (RR
51

The Parsing Process
Remaining Input:
))()
Next
symbol
B -> (RB | ε
Steps of leftmost
derivation:
B
(RB
((RRB
R -> ) | (RR
52

The Parsing Process
Remaining Input:
)()
Next
symbol
B -> (RB | ε
Steps of leftmost
derivation:
B
(RB
((RRB
(()RB
R -> ) | (RR
53

The Parsing Process
Remaining Input:
()
Next
symbol
B -> (RB | ε
Steps of leftmost
derivation:
B
(RB
((RRB
(()RB
(())B
R -> ) | (RR
54

The Parsing Process
Remaining Input:
)
Next
symbol
B -> (RB | ε
Steps of leftmost
derivation:
B
(())(RB
(RB
((RRB
(()RB
(())B
R -> ) | (RR
55

The Parsing Process
Remaining Input:
Next
symbol
B -> (RB | ε
Steps of leftmost
derivation:
B
(())(RB
(RB
(())()B
((RRB
(()RB
(())B
R -> ) | (RR
56

The Parsing Process
Remaining Input:
Next
symbol
B -> (RB | ε
Steps of leftmost
derivation:
B
(())(RB
(RB
(())()B
((RRB
(())()
(()RB
(())B
R -> ) | (RR
57

LL(1) Grammars

As an aside, a grammar such B -> (RB | ε
R -> ) | (RR, where you can always figure out the
production to use in a leftmost derivation by scanning
the given string left-to-right and looking only at the
next one symbol is called LL(1).

“Leftmost derivation, left-to-right scan, one symbol of
lookahead.”
58

LL(1) Grammars – (2)


Most programming languages have LL(1)
grammars.
LL(1) grammars are never ambiguous.
59

Inherent Ambiguity


It would be nice if for every ambiguous grammar,
there were some way to “fix” the ambiguity, as
we did for the balanced-parentheses grammar.
Unfortunately, certain CFL’s are inherently
ambiguous, meaning that every grammar for the
language is ambiguous.
60

Example: Inherent Ambiguity


The language {0i1j2k | i = j or j = k} is inherently
ambiguous.
Intuitively, at least some of the strings of the form
0n1n2n must be generated by two different parse
trees, one based on checking the 0’s and 1’s, the
other based on checking the 1’s and 2’s.
61

One Possible Ambiguous
Grammar
S -> AB | CD
A -> 0A1 | 01
B -> 2B | 2
C -> 0C | 0
D -> 1D2 | 12
A generates equal 0’s and 1’s
B generates any number of 2’s
C generates any number of 0’s
D generates equal 1’s and 2’s
And there are two derivations of every string
with equal numbers of 0’s, 1’s, and 2’s. E.g.:
S => AB => 01B =>012
S => CD => 0D => 012
62

Normal Forms for CFG’s
Eliminating Useless Variables
Removing Epsilon
Removing Unit Productions
Chomsky Normal Form
63

Variables That Derive Nothing


Consider: S -> AB, A -> aA | a, B -> AB
Although A derives all strings of a’s, B derives no
terminal strings.


Why? The only production for B leaves a B in the
sentential form.
Thus, S derives nothing, and the language is empty.
64

Discovery Algorithms




There is a family of algorithms that work
inductively.
They start discovering some facts that are obvious
(the basis).
They discover more facts from what they already
have discovered (induction).
Eventually, nothing more can be discovered, and
we are done.
65

Picture of Discovery
And so on …
Start with
the basis
facts
Round 1:
Add facts
that follow
from the
basis
Round 2:
Add facts
that follow
from round 1
and the
basis
66

Testing Whether a Variable
Derives Some Terminal String


Basis: If there is a production A -> w, where w has
no variables, then A derives a terminal string.
Induction: If there is a production A -> , where 
consists only of terminals and variables known to
derive a terminal string, then A derives a terminal
string.
67

Testing – (2)



Eventually, we can find no more variables.
An easy induction on the order in which variables
are discovered shows that each one truly derives a
terminal string.
Conversely, any variable that derives a terminal
string will be discovered by this algorithm.
68

Proof of Converse



The proof is an induction on the height of the leastheight parse tree by which a variable A derives a
terminal string.
Basis: Height = 1. Tree looks like:
A
Then the basis of the algorithm
tells us that A will be discovered.
...
a
1
69

an
Induction for Converse



Assume IH for parse trees of height < h, and
suppose A derives a terminal string via a parse
tree of height h:
By IH, those Xi’s that are variables are
discovered.
Thus, A will also be discovered, because it has a
right side of terminals and/or discovered
A
variables.
X1
w1
...
Xn
wn
Algorithm to Eliminate
Variables That Derive Nothing
1.
2.
Discover all variables that derive terminal
strings.
For all other variables, remove all productions in
which they appear in either the head or body.
71

Example: Eliminate Variables
S -> AB | C, A -> aA | a, B -> bB, C -> c

Basis: A and C are discovered because of A -> a
and C -> c.

Induction: S is discovered because of S -> C.

Nothing else can be discovered.

Result: S -> C, A -> aA | a, C -> c
72

Unreachable Symbols



Another way a terminal or variable deserves to be
eliminated is if it cannot appear in any derivation
from the start symbol.
Basis: We can reach S (the start symbol).
Induction: if we can reach A, and there is a
production A -> , then we can reach all symbols of
.
73

Unreachable Symbols – (2)


Easy inductions in both directions show that when
we can discover no more symbols, then we have all
and only the symbols that appear in derivations from
S.
Algorithm: Remove from the grammar all symbols
not discovered reachable from S and all productions
that involve these symbols.
74

Eliminating Useless Symbols
A symbol is useful if it appears in some derivation
of some terminal string from the start symbol.
Otherwise, it is useless.
Eliminate all useless symbols by:


1.
2.
Eliminate symbols that derive no terminal string.
Eliminate unreachable symbols.
75

Example: Useless Symbols – (2)


S -> AB, A -> C, C -> c, B -> bB
If we eliminated unreachable symbols first, we
would find everything is reachable.
A, C, and c would never get eliminated.
76

Why It Works



After step (1), every symbol remaining derives some
terminal string.
After step (2) the only symbols remaining are all
derivable from S.
In addition, they still derive a terminal string,
because such a derivation can only involve symbols
reachable from S.
77

Epsilon Productions

We can almost avoid using productions of the form A
-> ε (called ε-productions ).


The problem is that ε cannot be in the language of any
grammar that has no ε–productions.
Theorem: If L is a CFL, then L-{ε} has a CFG with
no ε-productions.
78

Nullable Symbols



To eliminate ε-productions, we first need to
discover the nullable symbols = variables A such
that A =>* ε.
Basis: If there is a production A -> ε, then A is
nullable.
Induction: If there is a production
A -> , and
all symbols of  are nullable, then A is nullable.
79

Example: Nullable Symbols



S -> AB, A -> aA | ε, B -> bB | A
Basis: A is nullable because of A -> ε.
Induction: B is nullable because of B -> A.
Then, S is nullable because of S -> AB.
80

Eliminating ε-Productions


Key idea: turn each production A -> X1…Xn into
a family of productions.
For each subset of nullable X’s, there is one
production with those eliminated from the right
side “in advance.”

Except, if all X’s are nullable (or the body was
empty to begin with), do not make a production with ε
as the right side.
81

Example: Eliminating εProductions
S -> ABC, A -> aA | ε, B -> bB | ε, C -> ε


A, B, C, and S are all nullable.
New grammar:
S -> ABC | AB | AC | BC | A | B | C
A -> aA | a
B -> bB | b
Note: C is now useless.
Eliminate its productions.
82

Why it Works
Prove that for all variables A:

1.
2.
If w  ε and A =>*old w, then A =>*new w.
If A =>*new w then w  ε and A =>*old w.

Then, letting A be the start symbol proves that
L(new) = L(old) – {ε}.

(1) is an induction on the number of steps by
which A derives w in the old grammar.
83

Proof of 1 – Basis



If the old derivation is one step, then A -> w must
be a production.
Since w  ε, this production also appears in the new
grammar.
Thus, A =>new w.
84

Proof of 1 – Induction



Let A =>*old w be a k-step derivation, and assume
the IH for derivations of fewer than k steps.
Let the first step be A =>old X1…Xn.
Then w can be broken into w = w1…wn, where Xi
=>*old wi, for all i, in fewer than k steps.
85

Induction – Continued


By the IH, if wi  ε, then Xi =>*new wi.
Also, the new grammar has a production with A
on the left, and just those Xi’s on the right such
that wi  ε.


Note: they can’t all be ε, because w  ε.
Follow a use of this production by the derivations
Xi =>*new wi to show that A derives w in the new
grammar.
86

Unit Productions




A unit production is one whose body consists of
exactly one variable.
These productions can be eliminated.
Key idea: If A =>* B by a series of unit
productions, and B ->  is a non-unit-production,
then add production A -> .
Then, drop all unit productions.
87

Unit Productions – (2)



Find all pairs (A, B) such that A =>* B by a
sequence of unit productions only.
Basis: Surely (A, A).
Induction: If we have found (A, B), and B -> C is a
unit production, then add (A, C).
88

Proof That We Find Exactly the
Right Pairs


By induction on the order in which pairs (A, B)
are found, we can show A =>* B by unit
productions.
Conversely, by induction on the number of steps
in the derivation by unit productions of A =>* B,
we can show that the pair (A, B) is discovered.
89

Proof The the Unit-ProductionElimination Algorithm Works


Basic idea: there is a leftmost derivation A =>*lm
w in the new grammar if and only if there is such
a derivation in the old.
A sequence of unit productions and a non-unit
production is collapsed into a single production of
the new grammar.
90

Cleaning Up a Grammar
Theorem: if L is a CFL, then there is a CFG for L –
{ε} that has:

2.
No useless symbols.
No ε-productions.
3.
No unit productions.
1.

I.e., every body is either a single terminal or has
length > 2.
91

Cleaning Up – (2)
Proof: Start with a CFG for L.
Perform the following steps in order:


1.
2.
3.
4.
Eliminate ε-productions.
Eliminate unit productions.
Eliminate variables that derive no terminal string.
Eliminate variables not reached from the start symbol.
Must be first. Can create
unit productions or useless
variables.
92

Chomsky Normal Form
A CFG is said to be in Chomsky Normal Form
if every production is of one of these two forms:

1.
2.

A -> BC (body is two variables).
A -> a (body is a single terminal).
Theorem: If L is a CFL, then L – {ε} has a CFG
in CNF.
93

Proof of CNF Theorem


Step 1: “Clean” the grammar, so every body is
either a single terminal or of length at least 2.
Step 2: For each body  a single terminal, make the
right side all variables.


For each terminal a create new variable Aa and
production Aa -> a.
Replace a by Aa in bodies of length > 2.
94

Example: Step 2


Consider production A -> BcDe.
We need variables Ac and Ae. with productions
Ac -> c and Ae -> e.


Note: you create at most one variable for each terminal,
and use it everywhere it is needed.
Replace A -> BcDe by A -> BAcDAe.
95

CNF Proof – Continued


Step 3: Break right sides longer than 2 into a chain
of productions with right sides of two variables.
Example: A -> BCDE is replaced by A -> BF, F ->
CG, and G -> DE.

F and G must be used nowhere else.
96

Example of Step 3 – Continued



Recall A -> BCDE is replaced by A -> BF, F ->
CG, and G -> DE.
In the new grammar, A => BF => BCG =>
BCDE.
More importantly: Once we choose to replace A
by BF, we must continue to BCG and BCDE.

Because F and G have only one production.
97


similar documents